Progress Report Working Area Data Assimilation - Prepared by: Period: Date
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Regional Cooperation for
Limited Area Modeling in Central Europe
Working Area Data Assimilation
Progress Report
Prepared by: Area Leader Benedikt Strajnar
Period: 2020
Date: 16/03/2020
1
Regional Cooperation for
Limited Area Modeling in Central Europe
Progress summary
The report summarizes the RC LACE DA activities in 2020.
The research and development work focused on application of radar assimilation in order to
enhance the knowledge of the underlying humidity pseudo-observation algorithm and to
investigate suitability of observation operator to be used with ALARO physics. Out of 3 planned
stays on radar DA, two were executed (partly influenced by Covid-19), and an additional stay
was focused on radar winds instead of the planned aircraft observations. Consequently, LACE
is now on a good track to reach satisfactory results with radar reflectivity. At the same time,
we would like to reach and provide a robust solution to radar dealiasing, to be able to use at
least part of radar Doppler winds. The stand-alone preprocessing software is meant for easier
handling and consolidation of those developments to be shared within members.
The Covid-19 epidemic caused a drop in AMDAR observations, the data type with very high
impact in DA systems run by LACE countries. To mitigate the drop in quality of forecasts,
EMADDC of KNMI made an expedited progress in collection in distributing additional Mode-S
observations. Their service now provides data from the MUAC area and additionally Denmark,
Austria, Slovenia and Romania, and is on the way to ingest more data over the LACE area. The
data was immediately shared/implemented with OPLACE, and some countries started to
benefit from this very large additional data counts. Progress is visible also on use of local
Mode-S data sources, although the situation with decreased data counts is not ideal for
evaluation of these data sets.
Increased attention was given to validation of SEKF and related surface observations, apart
from continuation of works in Austria the impact of SEKF was also substantially studied in
Hungary, on several periods and interesting weather cases. The results indicate visible
improvement in the description of the daily 2 m temperature cycle. Assimilation of surface
data in the currently operational OI was tuned in some member countries.
There was a number of validation and impact studies for a variety of observations, many of
which resulted in operational observation upgrades, as indicated in the summary tables.
Even more than usual, considerable part of the DA efforts were devoted to maintenance and
upgrades in the local DA systems. In a large part, those efforts were invested into migration
towards cy43 and accompanying modernization of DA suites. Members registered more than
15 person/months (out of planned 10 for the whole year) to accomplish this mostly technical
work. These mainly local DA efforts are summarized in the first section of this report.
2
Regional Cooperation for
Limited Area Modeling in Central Europe
Action/Subject/Deliverable: Maintenance and evolution of state-of-the-art DA systems [COM 3,
DA 1.6]
Description and objectives:
An overview of the current operational DA systems in RC LACE countries are presented in the
following two tables (yellow colors indicate the system upgrades and additions made in 2020
and first months of 2021 for cycle upgrades):
DA AUSTRIA AUSTRIA C- CROATIA AL- CZECH REP. HUNGARY HUNGARY SLOVAKIA AL- SLOVENIA AL-
AROME LAEF ARO ALARO ALARO AROME ARO ARO
Resol 2.5L90, 2.5L90, 4.0L73 2.3L87-NH 8L49 2.5L60 4.5L63 4.4L87
600 x 432 600 x 432 480 x 432 1069 x 853 432 x 432
Cycle 40t1 40t1 38t1_bf8 43t2pt_op1 cy43t2bf11 cy43t2bf11 cy43t2bf11 43t2_bf10
LBC IFS 1h IFS-EPS IFS 3h ARP 3h IFS 3h IFS 1h ARP 3h IFS 1h/3h
(lagged) (lagged) (lagged) (lagged) (lagged)
Method OI_main MES- OI_main MES- OI + 3D-Var OI + BlendVar OI + 3D-Var OI_main + OI + DF Blend- OI + 3D-Var
CAN + 3d-Var CAN + 3d-Var, 3D-Var ing
pert. obs. + Jk
Cycling 3h 6h 3h 6h 6h 3h 6h 3h
B matrix Downscaled static C-LAEF NMC method EDA EDA EDA - Downscaled
LAEF 11 km EDA ECMWF ENS
Initiali- No (SCC) No (SCC) No (SCC) IDFI in pro- DFI No No No (SCC)
zation duction, SCC
Obs. Synop + AS Synop + AS Synop Synop + AS Synop + AS Synop + AS Synop + AS Synop + AS
Amdar Amdar Amdar/MRAR (soil) Amdar GNSS ZTD Amdar/MRAR
Geowind Geowind Geowind Amdar/MRAR Geowind Amdar/MRAR / EHS
Temp Temp, Temp /EHS Temp Seviri Temp, Geowind
ASCAT, ASCAT, Seviri Ge- AMSUA/MHS Mode-S Temp Seviri
Snow- Snow- owind/HRWI ASCAT MRAR (SI, CZ), AMSUA/MHS
grid/MODIS grid/MODIS ND,Profiler, /IASI
snow- ASCAT, ASCAT/OSCAT
mask.,Mode- Temp E-GVAP ZTD
S EHS Seviri, (passive)
EMADDC
Table E1: Operational DA for NWP systems run by RC LACE countries.
DA AUSTRIA AROME-RUC CZECH REP. VarCanPack
Resol 1.2 L90 900 x 576 2.3L87-NH 1069 x 853
Cycle 40t1 43t2pt_op1
LBC AROME 1h -
Method OI_main MESCAN + 3d-Var + LHN + FDDA 3DVAR + OI
Cycling 1h -
B matrix Static EDA + differences of the day EDA
Initialization IAU -
Obs. Synop + AS, Amdar/MRAR/EHS national, EHS EMADDC, Synop + AS, Amdar/MRAR/EHS, Geowind/HRWIND,Pro-
Geowind, Temp, Seviri, AMSUA/MHS/HIRS/ATMS/IASI (+ filer, ASCAT, Seviri
Metop-C), ASCAT, GNSS ZTD (Austria), GPSRO (OPLACE),
Radar RH/Dow, INCA + AS at hig.freq., MODIS snowmask
3
Regional Cooperation for
Limited Area Modeling in Central Europe
Table E2: Operational DA for NWP-based systems nowcasting systems at hourly scale run by
RC LACE countries.
In Austria, cy43t2 export was compiled and all local adaptations from ZAMG added (Jk,
Nudging radar saturation, etc.). All steps were successfully technically tested and it was
checked that local modifications do not interfere with other development. Parallel runs of
AROME-Aut, AROME-RUC and CLAEF based completely on cy43t2 are currently running at
ZAMG/ECMWF and are expected to become operational within this year. The domains are
unchanged compared to oper versions except that AROME-AUT uses now exactly the same
domain as CLAEF. The orography switched from GTOPO to GMTED with one filtering in PGD
step. For AROME-RUC orography is unchanged (SRTM). Also the 2m diagnostics changed: In
CLAEF/AROME-AUT CANOPY-scheme is still active but modified, in AROME-RUC Dian 2016
diagnostics (without CANOPY) is applied. Regarding DA, REDNMC switch is reduced drastically
in AROME-Aut-parallel from 1.2 to 0.5 (as it was found to cause too dry near ground conditions
in the mountains). As the number of AMDAR reports dropped drastically during the Covid-19
crisis. Additional EMADDC MODE-S data (including MUAC, DK, AT) was added to all three
AROME systems in spring 2020.
In Croatia, the operational suite did not change. The plans for this year were heavily affected
by earthquake in Zagreb, so the priority was to set up backup of operations at ECMWF HPC, so
many efforts were devoted to that, together with increased operational maintenance on local
side. DA system with cy43 was set up in the e-suite which is planned for operations after
migration to the new HPC (to be available in the first half of 2021).
In the Czech Republic, local efforts were dedicated to impact studies for several observation
types: as outcome, high resolution wind atmospheric motion vectors (HRWIND), wind
profilers, SCAT are now assimilated by BlendVar and national SYNOP stations are used for soil
analysis (after sensitivity study with correlation functions in CANARI). Based on research in
2019 (visit), a full EDA B-matrix was operationally implemented. An hourly (non-cycled)
diagnostic analysis system VarCanPack is now operational (3D-VAR with short cut-off and OI
for screen level parameters). Czech DA team was very involved in radar assimilation validation
(2 stays and local work) as described later in the report.
In Hungary, the use of Mode-S observations (from Czech Republic and Hungary) and wind
profiler data were investigated. For soil assimilation, the SEKF was tested within AROME-
SURFEX for several periods and important cases when the quality of operational forecast was
insufficient. The new 90 level EDA-based B-matrix was evaluated and Desrozier’s method was
applied to search for an optimal tuning of observations and background errors. Use of Mode-
4
Regional Cooperation for
Limited Area Modeling in Central Europe
S and AMV observations was investigated. A switch to cycle cy43t2_bf11 was made in March
2021.
In Slovakia, efforts were put into porting and validation of data assimilation configurations in
CY43t2 (CANARI, DFI blending, 3D-Var configurations) and the introduction of BlendVar. Both
single observation experiments and full observations experiments have been evaluated. Some
of the fixes have been promoted for CY43t2bf11. The switch to operations (for blending) was
done in January 2021. The scripting system for DA assimilation (Perl scripts) was redefined in
terms of modules, VARBC for GNSS was also implemented. Considerable efforts were spent
on various tuition on DA topics. A nowcasting system prototype at 2 km was designed and first
tests with surface assimilation were performed. A QC system for surface stations based on
LAEF-A is under design. An examination of TEMP data in BUFR format was performed.
In Slovenia, most local work was devoted into design of a new nowcasting NWP setup (called
NWCRUC) with 1.3 km horizontal resolution and hourly cycling. The system was run regularly
over most of 2020 and was discontinued later in autumn to enable more validation runs as
described below. An impact study of OSCAT was carried out and this led to operational
implementation. By the end of the year, a change in the DA assimilation sequence (such that
Blendsur step was moved the between surface and upper-air analysis step) was developed and
was put to operations in early 2021. At the same time, the ODB_IO_METHOD=4 was
introduced in order to reduce the number of pool subdirectories in the ODB. Considerable
improvements in verification packages (Harmonie monitor and harp) were achieved by adding
the automatic stations from Italy, Austria and Slovenia to the surface verification datasets. First
steps towards installing the Obsmon DA validation package were made.
In Romania, the assimilation activities have been frozen in 2020.
In Poland, the activities towards the first assimilation suite are ongoing, the model cycle
43t2_bf10 will be used.
Action/Subject/Deliverable: Basic data assimilation setup (DAsKIT) [SPDA 1]
Description and objectives:
DAsKIT currently defines a basic set of DA infrastructure to ease developments of the DA
from consortium members with no prior experience. Polish DA development is so far
coordinated within the DAsKIT. Work included migration of existing 6 h DA cycle with CANARI
from cy40t1 to cy43t2.
Efforts: 3 months
Contributors: M. Szczech-Gajewska (Pl) 3
5
Regional Cooperation for
Limited Area Modeling in Central Europe
Documentation: /
Status: ONGOING
Action/Subject/Deliverable: Development of assimilation systems suited for nowcasting (RUC,
RAP, cycled and non-cycled hourly DA systems) [DA 5.2, DA 5.5, DA 7.2]
Description and objectives:
Several implementations of rapid-update cycle are under development or evaluation, and
two of the system reached operational status.
The AROME-RUC of Austria was put to operations by the end of 2019 but is under
constant evaluation and improving since then. The order of CANARI and 3D-Var step was
changed from parallel to usage of CANARI output in 3D-Var. For the latter, the 2m analyzed
values are copied to 3D-Var first guess. Also the snow and surface temperature fields are
replaced with CANARI output. A new Blendsur-like routine was written collecting surface
temperature information from land, sea, inland water and combine it to
SURFTEMPERATURE according to the parts in the grid box. As T2m and RH2m is used in
both CANARI and 3D-Var an inconsistency of soil and atmosphere (double use of
increment) is avoided by the modification. The 2 m diagnostics in AROME-RUC changed
modified, in AROME-RUC the Dian (2016) diagnostics without CANOPY is applied. Much
effort was put on this as the 2 m values with the previous setting were not satisfying at all
with strong warm bias especially in stable conditions. In AROME-RUC-parallel BATOR for
RADAR switched from Harmonie to MF version causing more Doppler winds and less
reflectivity data entering the system as in cy40t1 (Harmonie version) and allows Hungarian
radars to enter the system. In the AROME-RUC parallel, bufrtemp data (profiles, but no
trajectories) is produced from a local database with an ASCII to bufr routine. This means a
much better vertical resolution of the observations (see impact on Fig. N1). The VARBC
cycling frequency combination script from HIRLAM was slightly adapted to VARBC format
changes in cy43t2. To start VARBC, coldstart with option 2 (set bias to first guess
departures) was chosen.
6
Regional Cooperation for
Limited Area Modeling in Central Europe
Figure N1: Radiosounding Munich 10868 3rd March 2021 06 UTC (left) and pseudosoundings from
AROME-RUC-parallel analysis with Bufrtemp (blue) and obsoul temp (orange) T/°C (middle), Q/g/kg
right.
In the Czech Republic, the VarCanPack nowcasting analysis framework was upgraded. The
system runs from 2016 but was revised this year. It is based on adaptation of the
operational NWP system ALADIN/CHMI. Only hourly analyses (0 hour forecast) are
considered as starting point, but the system is planned to be further extended by +6h
forecast. The 3DVAR and OI algorithm are employed and the operational forecast with
different lead times are used as the first guess, see Figure H1. The 3DVAR analysis is done
first, the hydrometeors are not analyzed but copied from the first guess. After 3D-Var the
OI method is used to analyze screen-level parameters (not soil as done elsewhere). The
observation cut-off time is reduced to 20 min to provide the hourly results as soon as
possible - approx. 30 min after validity time. TH system uses all observations assimilated
by ALARO BlendVar, except TEMPs which are not available in this very short cut-off time.
Around 2500 observations are assimilated in the upper-air analysis in the day-time, with
one half of it coming from aircraft observations. When Mode-S MRAR observations are
added on top of AMDAR, the observations counts increase by further 50%.
Figure H1: Design of VarCanPack analysis system at CHMI. First guesses for non-cycled assimilation
are from operational BlendVar.
In Slovenia, a 3D-Var based nowcasting system at 1.3 km horizontal resolution has been
set up and run regularly from June till November 2020 for evaluation purposes. The hourly
analysis using 3D-Var and OI for soil is carried out with a cut off time of 35 minutes after the
nominal analysis time. This is followed by a production forecasts up to 36 h, which provide a
selection of meteorological variables every 5 min (e.g. simulated radar reflectivity, Fig. H2) on top
of the hourly model output. This forecast was run every 3h and it is planned to increase
7
Regional Cooperation for
Limited Area Modeling in Central Europe
this to every hour in the preoperational phase. The analysis step is repeated 35 min later
(70 min after the nominal analysis time), and this second analysis is used to compute the
first guess for the next production run. The observation data set includes all types of the
ALADIN 4.4 km plus the radar reflectivity. Observations, such as Mode-S datasets, are used
with a higher spatial resolution. Apart from daily runs which were assessed subjectively by
the NWP department, an objective verification period was carried out for a summer and
winter period. For these two periods, four experiments were prepared: both reference
operational ALADIN 4.4 km setup (3h DA cycling) and the new hourly NWCRUC (1h cycling)
were rerun twice, with and without radar reflectivity data assimilation (see also the radar
section of this report).
Figure H2: (left) NWCRUC model domain with orography and (right) simulated radar reflectivity for
one precipitation case.
8
Regional Cooperation for
Limited Area Modeling in Central Europe
Figure H3: Verification scorecard, comparing NWCRUC and ALADIN 4.4 km over the summer
period.
The 36h forecasts were verified against Synop stations over the whole domain plus
automatic surface stations in Slovenia, Austria and Italy, and radiosondes. A comparison
between reference operational ALADIN 4.4 km and NWCRUC (both without radar data
assimilation) for the summer period is shown in Fig. H3 as a scorecard. The 1.3 km suite
significantly improves the near surface variables. A concern remains with cloudiness which is
degraded over the first 12 h. The impact on precipitation forecasts was further assessed
using the categorical verification. Fig. XXX4 of the radar section shows the frequency bias of 1
h precipitation sums over the winter period, for various thresholds. It is evident that all
models tend to overestimate the precipitation amounts for precipitation intensities below 5
mm/h. The 1.3 km runs however decrease this bias for about one half in comparison to 4.4
km model. Spatial verification is also planned.
Efforts: 6.5 months
Contributors: F. Meier (At) 3.75, P. Scheffknecht 1, B. Strajnar (Si) 1.75, A. Bučánek, A.
Trojáková (Cz)
Documentation: /
Status: ONGOING
9
Regional Cooperation for
Limited Area Modeling in Central Europe
Action/Subject/Deliverable: Studies of background error statistics in 3D-Var [DA 1.2 – 1.4, 2.5]
Description and objectives:
In Slovenia, an experimental EDA system with 20 ensemble members was developed at a
current horizontal resolution of 4.4 km in 2019. The ensemble is used to compute
background error covariances on daily basis (with the sample size of 160 differences over the
last 8 analysis times. Analysis of daily covariances show that in all cases the variances were
higher and more centered towards small scales in the EDA experiment, compared to
operational downscaled (spin-up) B-matrix. Sensitivity experiment was carried out over 1-
month period, using the operational, mean EDA and flow-dependent EDA. The results
indicated slightly detrimental impact of the EDA B-matrices on most of parameters except
short-range precipitation. Differences between mean and daily variances were relatively
smaller. It is expected that the tuning of the REDNMC parameter would be needed in further
evaluations. This work on EDA may be latter combined with development of an EnVar under
OOPS.
Figure B1: Daily evolution of divergence spectra in EDA at model level 63/87 over one month period,
and comparison with the downscaled B (OPER ALADIN/SI at 4.4 km).
In Hungary, efforts to recalculate B-matrix continued by running a full EDA experiment based
on the previously calculated B-matrix (from the downscaled EDA) and computing a new B-
matrix for 90 level model setup. An experiment over a longer period was prepared with the
new B-matrix with mixed results, so a retuning of DA was considered. The following tuning
10Regional Cooperation for
Limited Area Modeling in Central Europe
parameters based on the Desroziers method were calculated for the system: REDNMC=1.26
and SIGMAO_COEF=0.71. The experiments showed degraded forecast scores with these tun-
ing values, so in the next step the following ideas were tested: no GNSS assimilation (the dif-
ferent model top may cause lower model-equivalent), switching off the CANOPY scheme (sim-
ilar to Météo-France), choosing a different time period (the first tests were made in a very
rainy spring period). The studied time period was from 9 to 31 July 2020 with mainly anticy-
clone in the beginning and upper-level cyclone afterwards which caused lots of showers and
thunderstorms in the second part of the period. Beside the reference run with 60 vertical lev-
els, a dynamical adaptation run at 90 levels was carried out. Figure B2 shows impact of the
tuning on 2 m temperature forecast. The EXP1, EXP2 and EXP3 are different tunings of the 90-
level B-matrix and the use of Canopy scheme, according to the table BT1. A positive impact on
bias and RMSE was shown with tunings EXP1 and EXP2.
Table BT1: Settings in the Hungarian 90 level B-matrix experiments.
Experi- Number of
CANOPY REDNMC SIGMAO_COEF REDNMC_Q
ment vertical levels
REF 60 on 1.2 0.9 -
EXP1 90 off 1.26 0.71 -
EXP2 90 off 1.2 0.9 1.67
EXP3 90 on 1.2 0.9 1.67
Figure B2: Bias (solid line) and RMSE (dashed line) of 2 m temperature forecasts in the 0 UTC runs from
9 to 31 July 2020. Black: AROME/HU at 60 levels; blue: dynamical adaption run at 90 levels; yellow, red
and green: assimilation runs at 90 levels with EXP1, EXP2 and EXP3 settings, respectively.
Figure B3 illustrates also the impact of the REDNMC_Q settings. With the higher value of RED-
NMC_Q the large bias and RMSE of the 3-hour precipitation decreased in the beginning of the
lead time. However, a separate case study showed improvements in the longer rain accumu-
lation with the new 90-level setting. Overall, the finer vertical resolution also shows mostly
positive impact on the surface parameters and more or less neutral impact in the upper-air.
11Regional Cooperation for
Limited Area Modeling in Central Europe
Figure B3: Bias (solid line) and RMSE (dashed line) of precipitation forecasts in the 0 UTC runs from 9
to 31 July 2020. Black: AROME/HU at 60 levels; blue: dynamical adaption run at 90 levels; yellow, red
and green: assimilation runs at 90 levels with EXP1, EXP2 and EXP3 settings, respectively.
In Croatia, the Jk method for including large scale information from global model in the process
of minimization was tested. The related V-matrix was computed from 16-member ECMWF
ensemble interpolated to ALADIN-HR4 domain. Differences were calculated over one month
period and using 6-h forecast for 00 and 12 UTC (N =480). The Jk term needs to be truncated
such that only large scales are kept from the host model. To determine the exact scale of
truncation, the procedure from Guidard and Ficher (2008) was adopted. It can be observed
that beyond k* ~ 25 (Fig. B4), two spectra begin to diverge (this number is slightly different for
other levels), and little energy is contained in mesoscale spectra of the host model. We thus
do not expect to derive any useful information from the host model at these scales. With the
aim to keep only the largest scales from host model, the truncation was set at k* = 15 (115
km). This value of the truncation was kept constant on all model levels for all variables and all
assimilation cycles.
Figure B4: Variance spectrum for temperature (left) and vorticity (right) at level 59 of the ALADIN-HR4
setup.
12Regional Cooperation for
Limited Area Modeling in Central Europe
Rather short period of 6 days was used to get first verification. Results indicate that Jk has
improved forecast for first ~12 hours, mainly for humidity. Still, a more relevant result will be
obtained by extending the verification period. One case where using Jk improved precipitation
forecast is shown in Fig. B5. Reference run simulated too narrow precipitation field on 6
October 2020, while Jk run predicted a bit broader precipitation area which was more in
accordance with radar measurements.
Figure B5: 6-hour accumulated precipitation from reference (left), run with Jk (middle) and radar
measurement (right).
Efforts: 3.5 months
Contributors: B. Strajnar (Si) 0.5, V. Homonnai (Hu) 2, A. Stanešić (Cr) 0.75, E. Keresturi (Cr)
0.25
Documentation: /
Status: ONGOING
Action/Subject/Deliverable: Surface Data Assimilation using Extended Kalman Filter [SU2, SU
3.4]
Description and objectives:
Validation of EKF has (re)started in Hungary by validation over the Hungarian domain with
AROME cy40t1 and SURFEX 7.3. Two cases were first studied in detail by conducting impact
studies with respect to AROME/HU OI-MAIN operational model. 3-hourly data assimilation
cycle was started two weeks before the events. Forcing required to offline SURFEX run were
coming from AROME inline forecasts at 9 m (radiation, precipitation, wind, humidity, pressure).
In the winter case, too much low clouds and corresponding under- forecast of daytime surface
temperature were observed. For this case, the analysis of 2 m temperature and low-level
cloudiness using SEKF did not really differ from the result of the operational run: similarly false
low-level cloud and overestimated temperature analysis were obtained. The analysis
increments and Jacobians of the observation operator were investigated. Figure S1 shows a
boxplot representations of the Jacobian elements for the case of January 8, 2020. The values
13Regional Cooperation for
Limited Area Modeling in Central Europe
have the lowest variation during daylight hours, especially at 12 and 15 UTC, and are closest
to 0, i.e., the effect of assimilation is the smallest at these times.
Figure S1: Boxplots for Jacobi matrix elements at different times of day (8 January 2020): dT2M/dWG2
(top left), dRH2M/dWG2 (top right), dT2M/dTG2 (bottom left), dRH2M/dTG2 (bottom right).
On the second case, 8 April, 2020, the daily temperature cycle was studied (Fig. S2). 3-hourly
data assimilation cycle was started two weeks before to get appropriate first guess for the
experiment, similar as in the winter-case study. This period was slightly warmer than usual and
unusually dry with almost no precipitation over the whole country. It is a typical problem in
AROME/HU that the minimum temperature is usually overestimated and maximum
temperature is underestimated in these long-live anticyclonic cases. The 2-metre temperature
analysis was inaccurate in AROME/HU OI-MAIN (Fig. S2 a). More accurate analysis and forecast
was provided by the SEKF experiment, as shown in comparison with SYNOP station Baja.
14Regional Cooperation for
Limited Area Modeling in Central Europe
a) b)
c) d)
Figure S2: Summer case SEKF vs. OI comparison: a) SYNOP observation, b) and c) analysis of AROME/HU
OI-MAIN and SEKF, respectively, at 0 UTC on 8 April 2020, d) analysis, forecasts and observations of T
2m at Baja meteorological station.
Further experimentation was carried on in the second half of 2020 after the implementation
of the AROME cy43t2_bf11 with SURFEX 8.0+ version. In addition to the standard binaries
(001, 927 etc.) the SURFEX OFFLINE and SODA binaries were also tested. Then two test runs
were performed and validated. The summer experiment lasted from 9 to 31 July 2020 (2-week
spin up from 25 June), the winter experiment lasted from 25 November to 17 December 2019
(spin up from 11 November). The 2 m temperature shows large improvement for the nighttime
hours and the large warm bias in the nights was reduced by SEKF with respect to the
operational OI-MAIN (Fig. S3). SEKF was able to improve the 2 m temperature analysis
throughout the whole period (Fig. S4, left), however, its daytime forecasts over Hungary do
not differ spectacularly from the OI-MAIN ones (Fig. S4, right).
15Regional Cooperation for
Limited Area Modeling in Central Europe
0 UTC runs 12 UTC runs
Figure S4: Bias (dashed line) and RMSE (solid line) of 2 m temperature forecasts in the 0 and 12 UTC
Analysis and observation 12-hour forecast and observation
Figure S5: Evolution of 2 m temperature analysis and 12h forecast in the 0 UTC runs and observations
(green) from 9 July to 31 July 2020. Blue: cy43 with EKF, orange: cy43 with OI-MAIN, red: cy40 with OI-
MAIN.
Further investigation was devoted into the soil water content – 2 m temperature relation. Dew
point forecast of SEKF suffered from cold bias most likely caused by the too dry soil. The 2 m
relative humidity verification shows very similar behavior as the 2 m temperature one:
improvement at night and deterioration during the day. The case study for 15 July 2020
indicated that the 2-metre temperature analysis was inaccurate using OI-MAIN (Fig. S6, left).
More accurate analysis was provided by SEKF, especially in the central part of the country (Fig.
S6, middle). At the same time the 12-hour forecast of SEKF resulted in an unrealistic warm
pattern over the south part of the domain (Fig. S7, middle). Soil moisture was drastically
decreased when applying SEKF as shown by checking water content in a specific point in this
area when the prevailing soil texture was sand.
OI-MAIN SEKF SYNOP
Figure S6: 2 m temperature analysis in AROME cy43 with OI-MAIN and SEKF, observations at 0 UTC on
15 July 2020.
16Regional Cooperation for
Limited Area Modeling in Central Europe
OI-MAIN SEKF SYNOP
Figure S7: 12-hour forecast of 2 m temperature in AROME cy43 with OI-MAIN and SEKF, observations
at 12 UTC on 15 July 2020.
The analysis increments for soil temperature and soil moisture in different analysis times,
summed for the whole period and for all grid points over the domain, is shown in Fig. S8. The
TG2 increment is bigger for the nighttime and smaller for the daytime periods in case of SEKF.
For OI-MAIN the TG2 increments are small and consistent. In contrast, the WG2 increments
are large for daytime and smaller for nighttime for both methods and are also higher in the
SKF.
SUM[ABS(A-G)] TG2 A-G SEKF_mod-CY43 SUM[ABS(A-G)] WG2 A-G SEKF_mod-CY43
A-G OI-CY43
8,00E+06 A-G OI-CY43
9,00E+03
7,00E+06
8,00E+03
6,00E+06 7,00E+03
5,00E+06 6,00E+03
5,00E+03
4,00E+06
4,00E+03
3,00E+06
3,00E+03
2,00E+06 2,00E+03
1,00E+06 1,00E+03
0,00E+00
0,00E+00 0 3 6 9 12 15 18 21
0 3 6 9 12 15 18 21
hour hour
Figure S8: Summed soil temperature (TG2) and soil moisture (WG2) increments for all of the grid points
over the domain from 9 July to 31 July 2020 for SEKF (blue) and OI-MAIN (orange).
Verification of 2 m temperature shows small but significant overall improvement in the winter
period. The cold bias detected with OI-MAIN was slightly reduced by SEKF. Neutral impact was
obtained for dew point and small positive effect can be seen for 2 m relative humidity. In spring
a parallel suite is going to start with SEKF and after several test periods it is planned to be
turned operational in 2021.
In Austria, SEKF experiment at ZAMG have been a combination of AROME CY40T1 with code
extensions of HIRLAM to activate the ISBA diffusion scheme for forecasts and a standalone
17Regional Cooperation for
Limited Area Modeling in Central Europe
version of SURFEX 8.1 for the soil assimilation. This solution is facing the problem that the
input/output files of SURFEX7.3 (in AROME) and 8.1 are not compatible. Thus it is planned to
update AROME to CY43T2 and switch from the ISBA force-restore (3-L) to the ISBA diffusion
(DIF) soil scheme. As a first step in this direction, a comparison of the two soil schemes is taking
place. For this purpose, AROME was computed from 07/2018 to 06/2020 with ISBA 3-L and
D95 snow scheme on the one hand and ISBA DIF and 3-L snow scheme on the other hand.
AROME is run with the operational Austrian namelist setting on a 2.5 km grid, is coupled to IFS
and no data assimilation is computed in this test run. The soil fields have been initialized as
similar as possible for the two test runs and soil fields are cycled, which means that the
+24hour forecast is used as initial file for the next forecast run. Forecasts of T2M have been
validated against SYNOP stations in Austria. With respect to the complex topography in the
model domain, stations have been grouped into flat land stations, stations in alpine valleys
and mountain top stations. Furthermore, seasonal impacts have been investigated. Overall it
can be stated that the diffusion scheme is resulting in warmer and drier conditions (T2M and
RH2M) on average, thus performing slightly better (RMSE closer to zero) for the domain
investigated (Fig. S9).
Figure 5: Bias (left) and RMSE (right) for force-restore (blue) and diffusion (green) soil scheme averaged
for one year (20190701-20200630) and all SYNOP stations in Austria. Forecast runs are starting at
00UTC and have a forecast range of 24 hours.
Efforts: 6.75 months
Contributors: H. Tóth 3.75, S. Schneider (At) 3, V. Tarjáni (Sk)
Documentation: /
Status: ONGOING
18Regional Cooperation for
Limited Area Modeling in Central Europe
Action/Subject/Deliverable: Observations suitable for soil schemes [SU 2.2.3, SU 2.5.2]
Description and objectives:
In Austria, experiments to investigate the impact of a local observation error for SCATSAR-SWI
on the forecast of screen level parameters have been finalized. While the assimilation
algorithm is working properly, the improvement on the forecast quality compared to the global
observation error is small and not significantly positive (Fig. SU1). Given a the large additional
effort necessary to define the local observation error (2 independent satellite data sets and
one model run, all of the available for at least one year, are necessary as input data) is not
arguable compared to the increase in forecast performance for an operational
implementation.
Figure SU1: Probability density functions of innovations di (left) and residuals dr (right) for all
experiments. Within each column, each quantity is shown for the soil layers WG1 to WG6 from top to
bottom. Zero values were omitted for the computations. eGlob is the experiment with a global
observation error, eLoc1 and eLoc2 are experiment with different local observation errors.
The approach to assimilate LST data with SEKF in AROME is working, but the impact is rather
small and not continuously positive. As the surface temperature is a rapidly changing
parameter both in space and time due to the missing storing capacity of the surface, it is
19Regional Cooperation for
Limited Area Modeling in Central Europe
difficult to achieve long-lasting changes in the model soil due to the assimilation. Based on the
combined MSG-Sentinel3-LST data set that has been developed at ZAMG, measurements are
available with a spatial spacing of 1 km and a temporal spacing of 15 minutes (in case of cloud
free conditions). Comparing different settings lead to the conclusion that a short OFFLINE
perturbation window (1 hour) and a rapid update (hourly LST assimilation) leads to the best
results. AROME forecasts have been computed once per day for up to +24 hours. Furthermore
it is beneficial to use the soil temperature of the uppermost six layers (0 to 60 cm) as control
variables in the SEKF (given short memory of near surface fields). The forecast performance
currently limits operational implementation.
Efforts: 8.5 months
Contributors: S. Schneider (At) 4.5, J. Vural (At) 4
Documentation: /
Status: ONGOING
Action/Subject/Deliverable: Assimilation of radiance observations in DA systems [DA 3.6]
Description and objectives:
Efforts: 0 months
Contributors:
Documentation: /
Status: ON HOLD (no manpower)
Action/Subject/Deliverable: Implementation of RADAR reflectivity and radial wind [DA 3.1]
Description and objectives:
RADAR reflectivity and radial wind observations are essential components of a mesoscale DA
system especially with increased analysis cycle frequency. It received more attention in 2019
and 2020 by several stays and regular DA group videoconferences. Three stays were devoted
to radar assimilation, first two into assimilation reflectivity and the last one to the Doppler
winds.
During a stay in Ljubljana, A. Trojáková investigated the use of the radar reflectivity observation
operator within ALARO model configuration, studying the effect of graupel in the ALARO
microphysics scheme on simulated radar reflectivity. A short tests have been performed. The
ALARO version with prognostic graupel provides very similar precipitation structures with
20Regional Cooperation for
Limited Area Modeling in Central Europe
slightly smaller intensities compared to the version without prognostic graupel. This
conclusion was based on a single case study. The prognostic graupel in ALARO are still
considered as a very fresh development which needs detailed evaluations in NWP context. The
graupel initialization to zero is necessary for the reflectivity observation operator even in case
of ALARO without prognostic graupel to avoid extremely high simulated reflectivity in radar
assimilation. The initialization of graupel in both cases should be better understood and
eventually fixed. It was found that prognostic graupel in ALARO microphysics has positive
effect on simulated reflectivity, namely by a small reduction of random error (STD) of
reflectivity innovations, see Fig R1. To make a reference to existing operational radar
assimilation, ALARO and AROME-FR reflectivity innovations were compared over Germany.
Both setups provide qualitatively similar statistics which give us more confidence for further
testing of radar reflectivity data assimilation within ALARO configuration.
Figure R1: Vertical profile of reflectivity innovations separated by DBZ-classes. Bias (left), standard
deviation (middle) and number of innovations for ALARO without graupel (red) and with graupel
(green).
The second stay in Vienna (Antonín Bučánek) was dedicated to familiarization with screening
procedure for radar reflectivity in AROME/ALARO, the construction of pseudo observed
relative humidity columns, and thinning. The work continued at CHMI after the stay has been
21Regional Cooperation for
Limited Area Modeling in Central Europe
shortened. Three minor bugs were spotted during investigation and discussed with MF. A
reproducibility problem was identified in thinning if 2 or more input hdf5 files from one station
were ingested. A few open questions related to the current data selection setup are still
opened. One is the modification in BATOR (in cy43t2_op4.03) for no-rainy observations as only
observations with minimum detectable signal ODB) within BATOR was briefly checked. Finally, passive data
assimilation experiments were prepared, observation-minus-background (OMG) statistics
were evaluated and first case studies were performed. The work continued by studying a few
significant cases. Fig. R2 shows a simulation of frontal precipitation band on 2020-05-23 with
and without assimilated Doppler wind (DOW). The differences are generally small.
Figure R2: Comparison of simulation using Doppler wind from all sites (left) and the reference
without DOW (middle), compared to the verifying OPERA composite.
Another experiment was to verify the impact of biased Danish dataset with the OPERA data.
Verification scores (6 h forecast) for all LACE OPERA radars (black) and all LACE OPERA radar
without DK and UK indicated improvement when omitting the two datasets.
In Austria, adaptations for cy43t2 were made in the prepopera preprocessing software
(addition of Nyquist velocity). The number of reflectivity assimilated is significantly lower
22Regional Cooperation for
Limited Area Modeling in Central Europe
within cy43t2 and for DOW slightly higher compared to cy40t1 probably because of presence
of check of QC flag > 0.7. The pre-processing was adapted that for stations where only 1 hour
old data were available in time now 15 min old data can be used – this is now the maximum
age for the AROME-RUC system. The problem of suspicious Doppler wind DFS values (orders
higher than other obs) was solved by re-definition of (hard coded) rejection limits (bugfix 11
cy43t2) and application of VARQC.
In Slovenia, three methods (the method used in Austria, CINDA method, torus mapping) for
wind dealiasing were implemented and tested on a set of precipitation cases over two years,
via check versus other observations (little overlapping observations) and more successfully
ALADIN/SI first guess. Figure R3 shows results of this OMG validation. The methods were
shown to give comparable results, the torus method being slightly better than others. The
torus mapping method has been chosen to be implemented in the HOOF preprocessing
system. Work on the reorganization of the package is ongoing.
Figure R3: OMG statistics for radar radial velocity after dealiasing. Elevations from 0 to 7 are
subject to de-aliasing, while upper 4 elevations (8-11). Reference (no de-aliasing) is shown in
red.
In the second part of 2020, a radar reflectivity impact study was carried out, in combination
with evaluation of the hourly NWCRUC system (see the Nowcasting application section). The
23Regional Cooperation for
Limited Area Modeling in Central Europe
experiment using reflectivity was compared with the reference run at both 4.4 and 1.2 km
resolutions. Radar sites from LACE countries (except Austria), Switzerland, Germany and
France were used, depending on the domain extent, with 3 h and 1 h cycle length and thinning
distances of 25 and 10 km at 4.4 and 1.3 km resolution, respectively. From subjective validation
it was concluded that the radar DA was especially capable of removing spurious precipitation
in the first guess. Additional moisture was added correctly to larger precipitation systems. For
smaller and isolated convection, the system often suffered from removing the rainy pixels by
the thinning procedure. Especially at kilometer resolution and hourly cycle, the DA system was
mostly able to follow the convection onset by correcting the hourly guess in the right direction,
as demonstrated in Fig. R4.
Figure R4: Simulated rainfall in one assimilation cycle (same end time) for the reference forecast (left
column), the run using radar reflectivity (middle) and the verifying INCA analysis (right). Top row: the
hourly NWCRUC system at 1.3 h, bottom row: the 3h ALADIN 4.4 km.
The OMG analysis over both periods indicate net drying of the atmosphere by the reflectivity
data assimilation. This is confirmed by the hourly precipitation verification (Fig. R5), using
Synop and automatic rainfall measurements from Slovenia, northeastern Italy and Austria.
Reduced frequency bias is apparent at both resolutions and this results in the increased ETS
for up to 6h into the forecast. The impact on other variables is more mixed (Fig. R5) and at 4.4
km resolution in general slightly negative in terms of bias (dry and cold bias) for upper air
fields. Standard deviation shows no significant changes. To still increase the impact of radar
observations, tuning of observation and background error is foreseen.
24Regional Cooperation for
Limited Area Modeling in Central Europe
Figure R5: Frequency bias (left) and ETS score (right) of hourly precipitation forecast. Runs are reference
NWCRUC (denoted nwc1), NWCRUC + radar reflectivity (nwr1), reference ALADIN 4.4 km (ref4) and
reference ALADIN 4.4 km + radar reflectivity (rad4). Period is 1-15 December 2020.
Figure R6: Scorecard of radar reflectivity experiments with respect to references for 1.3 km (left) and
4.4 km (right) resolutions in the experiment names are as in Fig R4.
In the Czech Republic, the first impact studies of reflectivity data assimilation was performed
but the main efforts were dedicated to spatial verification of precipitation. The idea is to com-
pute Fraction Skill Score (FSS) of model rainfall forecast against radar rainfall estimates + rain
25Regional Cooperation for
Limited Area Modeling in Central Europe
gauges product (called Merge2) over the Czech Republic. The verification is realized in R, the
FSS algorithm from verification package was optimized for better performance. The optimized
package is called fsspkg. When both model and observations are on the same grid the FSS is
computed for several box sizes and several precipitation thresholds.
Last but not least, the HOOF tool for homogenization of OPERA/OIFS radar measurements was
improved with a number of small corrections based on user feedback (discovered especially
during radar stays). Further work is needed for data splitting, which is not satisfactory for the
radar sites with irregular/random scanning order.
Efforts: 19 months
Contributors: A. Trojáková (Cz) 3.5, A. Bučánek (Cz) 6, V. Švagelj (Si) 2.5, B. Strajnar (Si) 2.5,
F. Meier(At) 0.5, K. Čatlošová (Sk) 3.5, M. Nestiak(Sk) 0.5
Documentation: Software and documentation/reports
Status: ONGOING
Action/Subject/Deliverable: Assimilation of GNSS path delays (ZTD, STD, GNSS-RO, refractivity
index, gradients) [DA 3.3, DA 4.2, DA 5.1]
Description and objectives:
The usage of GNSS-derived observations is planned in most of the RC LACE member countries
and is currently operational in Hungary and Austria (in AROME-RUC).
In Slovakia, a comparison between TPW (total precipitable water) from NWC SAF and PWV
(precipitable water vapour) from GNSS was performed. The PWV were computed from ZTD
estimated at near real time and atmospheric parameters from operational configuration
ALADIN/SK. Positive correlations were noticed between TPW and PWV at multiple GNSS
stations (Fig. G1). The assimilation tests of available STDs from GNSS station BASV were
performed, the positive increments (more specific humidity in analysis) are located mainly
eastwards of the station. Actually, some negative increments were noticed west of the station.
Asymmetrical increments at GNSS station are an expected feature of assimilation of all
available slant total delays at GNSS station. The STD code was compiled into cy46t1 on beaufix
(MF). The aim is to implement all the developments first to this release and later to cy48.
In Croatia, the ROM SAF GBGP (Ground-Based GNSS package) has been adapted and installed
on the local HPC to test the conversion from "COST" ASCII to BUFR format. For successful
installation, the default ifort compiler has to be replaced by gcc (8.2). Data conversion was
tested on an hourly operational data from E-GVAP by a GBGP "cost2bufr" function. A relevant
26Regional Cooperation for
Limited Area Modeling in Central Europe
entry for the param.cfg namelist of BATOR was prepared. Data conversion was tested on the
hourly operational data from E-GVAP by a GBGP "cost2bufr" function.
In Slovenia, it was agreed with data provider (GIS institute) to verify their observations by
comparison to other E-GVAP data, so it adapted the local solution, called SIGNAL. Initial
updated datasets were provided by the end of 2019 and now much more correspond to
other E-GVAP solution, compared on common reference GNSS sites. An evaluation of
those observations were performed in ALADIN, suing 3 VarBC predictors. In the first place,
comparison of observation and the model reveals systematic biases between SIGNAL2
and ALADIN. These can originate both from the model deficiencies or observations, and a
common source of such bias is a mismatch between station height and the model terrain
height. This is clearly visible for stations located in narrow valleys like BOVC (Bovec,
northwest Slovenia) which also gets one of the highest bias correction in the described
experiments. A one-month data set was shown to be too short for VarBC warm-up, so the
experiment was recycled for another month using the VarBC parameters from previous
run. Verification scores were computed for the experiments with 1 and 2 months of
warm-up cycling, against the operational setting for reference. The verification suggests
rather high impact of ZTD but increase of bias for 2 m temperature and humidity (reduced
in the experiment with longer warm-up period) while the cloudiness as well as some
layers in the free atmosphere are improved. Even though the results with ZTD assimilation
are not yet satisfactory, the new SIGNAL dataset performed clearly better than the
previously used dataset. The bias correction investigation suggests that a longer data set
will be needed to reach optimal observation values for ALADIN/SI (3-6 months). The final
agreement foresees provision of raw GNSS data to ARSO and the ZTD computation is to
be done with an external partner.
27Regional Cooperation for
Limited Area Modeling in Central Europe
Figure G: Evolution of bias estimates for Slovenian SIGNAL observations at 12 over the warm-up period.
By the end of 2020, E-GVAP observations were implemented into the OPLACE preprocessing
system. A whitelist of E-GVAP GNSS sites was computed over 30 days for Slovenian 4 .4 km
domain, and that included 930 stations (Fig. G2). Further work included modification of the
screening code (cy43t2) in order to use passive assimilation to enable a long-enough spin-up
time. Modifications also included a tool (based on HIRLAM solution and adapted by Florian
Meier) to merge hourly VARBC files for ZTD and daily files for satellite radiances. All this
developments were tested and prepared to enter the operations in 2021.
28Regional Cooperation for
Limited Area Modeling in Central Europe
Figure G2: Whitelisted (red) E-GVAP stations for the ALADIN/SI 4.4 km model domain.
Efforts: 3.75 months
Contributors: M. Imrišek (Sk), 2.5, B. Strajnar (Si) 1, S. Panežić (Cr) 0.25
Documentation: /
Status: ONGOING
Action/Subject/Deliverable: Assimilation of Mode-S (EHS and MRAR) observations [DA 3.2, DA
5.1]
Description and objectives:
The use of Mode-S high-resolution aircraft observations (both MRAR and EHS sub types) are
fast growing observations network with increasing importance in mesoscale DA systems.
During the Covid-19 outbreak in 2020, the EMADDC service of KNMI started to provide new
Mode-S data sets, containing EHS data from MUAC and Denmark, and later also Austria and
Romania (the so-called “test” and “test 2” data from secondary processing servers - not to be
interchanged by the previous MUAC data). In OPLACE, two data streams are now available as
“ehs_nl” (standard operational MUAC) and “ehs_eu” (EMADDC test 2, including data from
Denmark, Austria, Romania). A short obs-minus-guess study was conducted immediately after
the release of new EMADDC data set. From this first evaluation of a degradation of wind from
the Austrian and Romanian airspace compared to the Netherlands and Germany was
observed. In autumn during the DA working days Siebren de Haan presented upgrades in
29Regional Cooperation for
Limited Area Modeling in Central Europe
EMADDC processing of Mode-S data. Therefore OMG statistics of the European Mode-S EHS
data (EHS_EU) have been re-checked for the two periods: 2-12 October and 13-28 October
2020. The OMG statistics (MEAN, STD) separated by type and areas are illustrated in Fig. M1.
The overall EHS_EU wind statistics are now comparable with AMDARs. The separation by areas
shows somewhat larger temperature STD for Romanian airspace which was reported back to
EMADDC. An impact study using European Mode-S data has been also performed in the Czech
Republic but evaluations are still ongoing.
Fig M1: Vertical profile of aircraft innovations separated by geographical areas. BIAS (left), STD (middle)
and number of innovations for all AMDAR data (red), MUAC area (green), UK area (blue), Romanian
area (light blue), Austrian and Slovenian area (brown) for period of 13-28 October 2020.
In all Austrian AROME suites, the Mode-S EHS from MUAC and Denmark (available as “test”
data) are assimilated from May 2020 on top of national EHS observations. In AROME-RUC, the
“test 2” EMADDC data are also used operationally. However, they come too late for the hourly
analysis and therefore MUAC oper data are used on top for this time, while “test 2” data are
used for the hour before (-90min-+30min window). The impact was not evaluated so far, but
number of observations in AROME-Aut with new data is higher than before COVID AMDAR
only data - except very low levels (Fig. M2).
30Regional Cooperation for
Limited Area Modeling in Central Europe
Figure M2: Increase of aircraft number assimilated in AROME-Aut 2.5 km during COVID 19 crisis on
various pressure levels. In mid-May, MODE-S EHS from KNMI-test (OPLACE) was activated. The loss of
AMDAR can be rather well compensated by MODE-S, but still little less observations are available
around 925 hPa.
In Hungary, a series of experiments were carried out in order to investigate the impact of Czech
Mode-S MRAR data on the quality of the forecast. They covered a winter (December 2019)
and a summer period (June 2020) and were based on the operational AROME/HU. Pointwise
verification was achieved for both periods using SYNOP and TEMP observations. The verifica-
tion results show little improvement for the winter period: scores are usually very close to the
operational ones, although a slight improvement can be detected in most cases. A remarkable,
albeit small positive impact can be seen regarding precipitation (Fig. M2), although it has to
be noted that there were only a few days with considerable amount of precipitation in this
specific period. Results on the summer period were impacted by Covid-19 traffic reduction. It
31Regional Cooperation for
Limited Area Modeling in Central Europe
can be concluded that the assimilation of Czech Mode-S data yields only slightly different re-
sults in the forecast. The impact of the Mode-S dataset was quite substantial for specific days,
at least for the summer period, but the difference is usually much smaller. The beneficial
impact on precipitation is also reflected in increase of ETS (Figs. M3, M4). Other surface
variables, however, show little to no improvement, and this is quite similar to the impact seen
in the winter experiment. The assimilation of the Czech Mode-S MRAR data turns to
operational in AROME/HU from March 2021 (already in cycle 43t2).
Figure M2: 12-hour forecast of 3-hour precipitation sum in the experiment using Mode-S data (left),
reference run (middle), and the radar data (right) at 12 UTC on 21 June 2020.
Figure M3: ETS score of 12-hourly precipitation in the 18 UTC runs from 1 to 19 December 2019. Red
line: experiment with Czech Mode-S data, black line: experiment with Hungarian Mode-S data, green
line: reference experiment without any Mode-S data.
32Regional Cooperation for
Limited Area Modeling in Central Europe
Figure M4: ETS score of 12-hourly precipitation in the 9 UTC runs from 1 to 30 June 2020. Red line:
experiment with Czech Mode-S data, black line: reference run.
Figure M5: Bias (solid line) and RMSE (dashed line) of cloud forecasts in the 0 UTC runs. Red:
experiment with assimilation of Czech Mode-S data, green: experiment with assimilation of Hungarian
Mode-S data, black: operational AROME/HU without any Mode-S data.
Apart from using Czech MRAR data, first test with 4 weeks of Hungarian data, provided by
Hungarian ATC (HungaroControl) were conducted. The first verification results showed that
the accuracy of the AROME/HU forecasts were not affected remarkable by the raw MRAR
dataset (see Figs. M3 and M5, comparison between green and black line). At nighttime,
differences are slightly bigger with respect to the ones at daytime but their magnitudes are
33Regional Cooperation for
Limited Area Modeling in Central Europe
not considerable. The results were compared also to the results of the experiment with Czech
Mode-S data. After the first test phase, HungaroControl expanded the dataset to cover around
4 moths. As next step, a whitelisting procedure was carried out with similar thresholds as those
applied in the other LACE Members which resulted in retaining 72 % temperature and 64 %
wind observations of the dataset. The follow-up impact study show slight difference in
comparison with the reference forecasts, which are more apparent in the 12 UTC runs when
there are more flights (Fig. M6). The Hungarian Mode-S MRAR data positively affect the 2 m
temperature and relative humidity analysis leading to tiny improvements in precipitation,
relative humidity, total cloud cover and wind gust forecasts, while in case of the other variables
the impact is rather neutral. This investigation will continue after the Covid-19 pandemic.
Total cloud cover 10 m wind gust
Relative humidity Hourly precipitation
Figure M6: Bias (solid line) and RMSE (dashed line) of 12-hour forecasts for total cloud cover,
10 m wind gust, 2 m relative humidity and precipitation in the 12 UTC runs from 1 to 18 De-
cember 2019. Red line: experiment with Hungarian Mode-S data, black line: reference run.
In Slovakia, a comparison of Mode-S EHS and Mode-S MRAR data originating from the same
aircrafts was performed. The dataset was retrieved from Buchtuv kopec (Czech Republic)
where both data types were available over the period of Jan-Feb 2018. Significant differences
in temperature measurements were observed. An illustration of OMG departures for active
EHS and MRAR data is depicted in Fig. M7. Further experiments included work on tuning the
thinning of Mode-S data (Fig. M8). Results are rather mixed and the added value of Mode-S
cannot be clearly demonstrated (the impact is slightly negative for temperature). Longer
verification periods are ongoing.
34Regional Cooperation for
Limited Area Modeling in Central Europe
Figure M7: EHS (left) and MRAR (right) OMG departures computed for a particular flight.
Figure M8: Verification of Mode-S thinning experiments: 25 km & 15 hPa (red), 5 km & 5 hPa (green)
and no Mode-S data (blue) for temperature wind speed and wind direction.
Efforts: 13.25 months
Contributors: F. Meier (At) 0.5, Gabriella Tóth (Hu) 2.75, Anikó Várkonyi (Hu) 0.75, B.
Strajnar (Si) 0.25, K. Čatlošová (Sk) 6.5, Kristóf Szanyi 2, A. Trojákova (Cz) 0.5
Action/Subject/Deliverable: The use of AMV products (Geowind, HRW and Multi-Metop) [DA
3.5]
Description and objectives:
An impact study was carried out in Hungary in order to evaluate AMV (GEOWIND and
HRWIND) in AROME/HU over two periods with frequent weather fronts and convection. In
the assimilation, AMVs were used in both long and short forecasts as the timelines of the
data allows it. The AMVs were generated form IR, VIS and WV channels. For the thinning
and blacklisting, the same settings as used in ALADIN/HU was adopted. In both periods,
35You can also read
